Using social network analysis for spam detection

Abstract

Content filtering is a popular approach to spam detection. It focuses on analysis of the message content to identify spam. In this paper, we evaluate the use of social network analysis measures to improve the performance of a content filtering model. By measuring the degree centrality of message transfer agents, we observed performance improvements for spam detection in repeated experiments; e.g. a 70% increase in the proportion of spam detected with a false positive rate of 0.1%. We were also able to use anomaly detection to identify mislabeled messages in a publicly available spam data set. Messages claiming unusually long paths between the sender's message transfer agent and the recipient's message transfer agent turned out to be spam. © Springer-Verlag Berlin Heidelberg 2010.